Seminar Schedule




MCSC 310


The Pursuit of Engagement

Megan Mocko


University of Florida

Abstract: In this talk, I will discuss various methods used to increase engagement and learning in an introductory statistics course. The original GAISE guidelines (ASA, 2005) and the updated GAISE 2016 guidelines (GAISE College Report ASA Revision Committee) included the guideline “Foster Active Learning”. The updated GAISE guidelines also included the emphasis “Teach statistics as an investigative process of problem solving and decision making”. In this talk, two examples inspired by these ideas will be given, including the creation of active learning labs in a collaboration space for a 2,000 student hybrid (online lecture and in person lab) Introductory Statistics course and a multi-part project using a virtual environment for a completely online course.

Refreshments immediately following seminar in Room 301M MSCS.




MSCS 310

Addressing uncertainty, ill-posedness, and high-dimensionality in optimization and variational inequality problems

Dr. Farzad Yousefian
Assistant Professor
OSU - Department of Industrial Engineering and Management

Abstract: A wide range of emerging applications in machine learning, signal processing, and multi-agent systems result in optimization or variational inequality problems. These problems are often complicated by uncertainty, ill-posedness, and/or high-dimensionality. In this talk, we present three algorithms on addressing these challenges. In the first part of the talk, motivated by multiuser noncooperative Nash games in stochastic regimes, we consider stochastic variational inequality problems (SVI) on semidefinite matrix spaces. To solve this class of problems, we develop a stochastic mirror descent method. In the second part of the talk, we consider ill-posed stochastic optimization problems. We consider a bilevel model, where the goal is to find an optimal solution that attains the minimum value of a regularizer and develop an iterative regularized stochastic mirror descent method. We present the implementation on a text classification application. In the third part of the talk, we consider high-dimensional ill-posed convex optimization problems and develop a randomized block coordinate iterative regularized gradient method. The performance of the algorithm for solving linear inverse problems in image deblurring is presented. 

Refreshments immediately following seminar in Room 301M MSCS.


Stillwater Translational Genomics Forum 2019 - genomic selection and breeding optimization

Oklahoma State- Wheat Improvement Team and Translational Genomics Laboratory
Iowa State- Dr. William Beavis
Michigan State- Dr. Gustavo de los Campos
University of Alberta- Resilient Forest Project (RES-FOR)
University of British Columbia- Resilient Forest Project (RES-FOR)

Breeding, consisting of selection, crossing and evaluation, is a series of decision-making process. Under the framework of genomic selection, breeding values of all candidates are estimated by summarizing genome-wide information. While accurate estimation of breeding values for un-typed individuals has been actively investigated, breeding decisions on which individuals to select and cross to form the improved population are still critical for the success of crop improvement, which is often defined by multiple, competing objectives. Considering limited institutional resource available, breeding decisions present a classic multi-objective optimization problem, where solutions are sought to maximize genetic gain, to balance genomic diversity and the genetic connectivity to breeding target, and finally to assure long-term sustainability for all breeding objectives.
Stillwater Translational Genomics Forum is initiated as a cross-link between research expertise and the leadership in industries and national and international programs. Recognizing limiting factors and risks involved in technology uptakes, the Forum is poised to offer an open, intellectual dialogue that strategically identifies opportunities to transfer knowledge, practice and technologies between research, policy and industry.

09:00 – 09:30  Welcome Remarks and Introduction

09:30 – 10:30  Updates on Predictive Analysis for Crop Improvement (led by Dr. de los Campos)

10:45 – 12:00  Discussion

12:00 – 13:15  Lunch

13:30 – 14:30  Breeding Optimization (led by Dr. Beavis)

14:45 – 15:45  Discussion        

16:00 – 17:00  Dr. de los Campos – seminar (348B NRC Roger E. Koeppe Seminar Room)

             18:00  Drink and Dinner at Charles’



MSCS 310

New Directions in Bayesian Sparse Signal Recovery

Jyotishka Datta
University of Arkansas 

Abstract: Global-local shrinkage priors have been established as the current state-of-the art inferential tool for sparse signal recovery as well as the default choice for handling non-linearity in paradoxical problems without a Bayesian answer. Despite these success stories, certain aspects of their behavior, such as, validity as a non-convex regularization method, performance in presence of correlated errors or adapting to unknown error distribution, remain unexplored. In this talk, I will offer insightful solutions to some of these open problems motivated by modern applications. In the first half, I will discuss the notions of theoretical optimality for sparse signal recovery using global-local shrinkage priors. In the second half, I will build a non-convex prior-penalty dual that offers the best of both Bayesian and frequentist worlds, by merging full uncertainty characterization with fast and direct mode exploration. We will also propose a formal Bayesian solution to two related problems: (i) adaptation to an unknown error distribution leading to bimodality of the posterior density and (ii) handling a misspecified model when the true generating distribution is heavy-tailed, producing outliers in a Gaussian framework. 

(This is a joint work with Anindya Bhadra, Nick Polson, and Brandon Willard.)



MSCS 310

Henry Han
OSU PhD Candidate 

> cancelled < 



MSCS 310

Title: Bayesian Multi-Environment Genome-wide Prediction with Weighted Kernel

With the continued advancement of next-generation sequencing technology, genomic selection (GS, Meuwissen et al. 2001), a selection of breeding individuals based on genome-wide molecular markers, is becoming popular in the breeding programs. The success of GS relies heavily on the accuracy of phenotypic prediction through high-dimensional genomic data. Since each plant responds to environment differently, the observed phenotypes of breeding population(s) are not necessarily consistent over environments. Therefore, it is challenging for plant breeders to select breeding plants based on single-environment genome-wide prediction model with the confidence that phenotypes are stable and superior across other environments. To overcome this, in this talk, I will present a Bayesian genome-wide prediction model that can jointly predict the performance of breeding population(s) across multiple environments. In addition, a weighted kernel, which takes the biological insight of the trait into account, will be presented for the proposed multi-environment genome-wide prediction model. To assess the prediction performance of the proposed model and method, Oklahoma doubled haploid wheat population and Canada interior spruce data, each has four different phenotypic traits, and across three different environments, are analyzed in this study. It has been shown that the improved prediction accuracy can be expected from multi-environment genome-wide prediction when the phenotypic correlation between environments is moderate to high. Besides, the proposed weighted kernel has shown its power to achieve a better prediction accuracy than the traditional Gaussian kernel.

Xiaowei Hu
OSU PhD Candidate 



MSCS 310

Benjamin Waller & Jeff Huang
OSU Masters Students